Test accuracy is modeled on 1000 hypothetical patients with suspected HIT. We assumed a disease prevalence of 7.9%, the same prevalence as observed in the validation cohort of Marchetti et al. For the Marchetti algorithm, we assumed that the 2.9% of patients determined to be unclassifiable by the algorithm would be treated empirically for HIT; those patients ultimately found to have HIT by the reference standard were therefore classified as true-positives whereas those ultimately found not to have HIT were classified as false-positives. For the ASH algorithm, we used a sensitivity and specificity of 0.921 and 0.542, respectively, for the 4Ts score and 0.98 and 0.85, respectively, for the immunoglobulin G (IgG)–specific ELISA, the same values that were used in the ASH 2018 guideline on HIT.⁴